opponent-modelinglisted

# Opponent Modeling Without history you must play GTO baseline. After ~30 hands per opponent, exploit their tendencies. ## Stat thresholds (6-max NLHE, rough) | Stat | Tight | Standard | Loose | |------|-------|----------|-------| | VPIP | < 20% | 20–28% | > 28% | | PFR | < 14% | 14–22% | > 22% | | 3-bet | < 5% | 5–9% | > 9% | | AF (postflop) | < 2 | 2–4 | > 4 | ## Player archetypes - **Nit** (VPIP < 16, PFR < 14): folds too much, bluff often. Continuation-bet relentlessly. - **TAG** (VPIP 18–24, PFR 14–22): solid. Don't bluff into them; value bet thin. - **LAG** (VPIP 25–32, PFR 22–28, AF > 3): aggressive. Trap with strong hands; light 3-bets work. - **Maniac** (VPIP > 35, PFR > 30, AF > 4): bluffs constantly. Value bet wide; call down lighter. - **Calling station** (VPIP > 30, AF < 1.5): never folds. Stop bluffing; value bet medium-strong. ## Process each decision 1. `opponent_database_query opponent_id=<id>` — get VPIP/PFR/3-bet/AF/WTSD. 2. If `confidence == "no_data"`, default to GTO via `gto_lookup`. 3. Otherwise classify archetype and adjust. ## After each hand If something surprising happened — `note_manager append`: ``` note_manager( action="append", opponent_id="player_b", observation_type="bluff" | "value" | "tendency", content="3-bet 7-2o from button, then triple-barreled bluff. Maniac confirmed.", hand_id="h_00042", ) ``` Notes accumulate. Read them next time with `note_manager read`.